PDPei Dong

Pei Dong

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Pei Dong has a total of 48 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Bilingual Benchmark for Evaluating Diagnostic Performance of Multimodal Large Language Models in Radiology (RadM-Bench): Evaluation Development and Validation

    by Qingxia Wu, Qingxia Wu, Peipei Zhang, Zhifeng Yi, Yu Shen, Yan Bai, Hongna Tan, Pei Dong, Zhong Xue, Neil Roberts, et al.

    Background: Multimodal large language models are increasingly used in radiological diagnosis, but their performance has not been systematically evaluated across volumetric (3D) imaging, real-world clinical versus public teaching cases, and bilingual contexts. Objective: The aim of the study is to develop a bilingual radiology benchmark and characterize the diagnostic performance of state-of-the-art multimodal large language models across input modality, clinical setting (public teaching vs routine clinical), disease rarity, and clinical-history language and to disentangle linguistic from clinical-content effects through a cross-linguistic control experiment. Methods: We constructed RadM-Bench, comprising 720 cases evenly distributed across 9 radiological subspecialties: 360 English public teaching cases enriched in rare diseases (RadEdu) and 360 Chinese routine clinical cases (RealClin). In total, 4 proprietary models (GPT-4o, O3, Gemini-2-Flash, and Gemini-2.5-Flash-Thinking) and 6 open-source models (Qwen2.5-VL-72B/7B, InternVL3-78B/8B, Llama-4-Scout-17B-16E, and MedGemma-4B) were evaluated under 4 input conditions: clinical history alone, history with radiologist-selected 2D key images, and history with volumetric data sampled at 2 and 10 frames per second (fps). Each response was scored on a 4-tier 0‐3 diagnostic-quality rubric by 2 board-certified radiologists blinded to model identity. Mean scores with bias-corrected and accelerated bootstrap 95% CIs are reported. To disentangle language from clinical content, all 360 RealClin histories were translated into English and re-evaluated, with paired comparisons by Wilcoxon signed-rank tests and Benjamini-Hochberg false-discovery-rate correction. Results: Mean performance remained below 1.5 on the 0‐3 scale for all 10 models on both datasets. Adding radiologist-selected 2D key images to clinical history improved performance in all 10 models (+19.8% to +139.2%). In RealClin at fps=10, all 8 evaluable models scored lower with volumetric input than with the 2D-image baseline (−5.3% to −31.4%); MedGemma-4B and Llama-4-Scout-17B-16E could only be evaluated at fps=2 due to context-window and graphics processing unit–memory constraints. At fps=2, a total of 8 out of 10 models declined (−6.8% to −28.6%), while Qwen2.5-VL-7B and InternVL3-8B showed marginal improvements (+2.4% and +2.1%). Cross-dataset transfer diverged by model category: proprietary models declined from RadEdu to RealClin (eg, O3 with images: 1.14 to 0.79), whereas Chinese-centric open-source models improved (eg, InternVL3-78B: 0.48 to 0.75). The rare-disease premium observed in 9 out of 10 models in RadEdu reversed in RealClin, where common-disease scores exceeded rare-disease scores in 7 out of 10 models under history-only input. Translating RealClin histories into English produced a numerical decrease in mean score for all 10 models, which were statistically significant in 9 out of 10 models after false discovery rate correction, excluding a Chinese-language penalty. Conclusions: Within the scope of this benchmark, multimodal inputs improved performance over clinical history alone, but performance gaps remain in volumetric data processing and cross-context generalization, with mean diagnostic performance across the 10 evaluated models remaining below clinically actionable levels on both datasets.

  • A topology-preserving three-stage framework for fully-connected coronary artery extraction

    by Yuehui Qiu, Dandan Shan, Yining Wang, Pei Dong, Dijia Wu, Xinnian Yang, Qingqi Hong, Dinggang Shen


  • Open Access

    Evaluating Large Language Models for Automated Reporting and Data Systems Categorization: Cross-Sectional Study

    by Qingxia Wu, Huali Li, Yan Wang, Yan Bai, Yaping Wu, Xuan Yu, Xiaodong Li, Pei Dong, Jon Xue, Dinggang Shen, et al.

    Background: Large language models show promise for improving radiology workflows, but their performance on structured radiological tasks such as Reporting and Data Systems (RADS) categorization remains unexplored. Objective: This study aims to evaluate 3 large language model chatbots—Claude-2, GPT-3.5, and GPT-4—on assigning RADS categories to radiology reports and assess the impact of different prompting strategies. Methods: This cross-sectional study compared 3 chatbots using 30 radiology reports (10 per RADS criteria), using a 3-level prompting strategy: zero-shot, few-shot, and guideline PDF-informed prompts. The cases were grounded in Liver Imaging Reporting & Data System (LI-RADS) version 2018, Lung CT (computed tomography) Screening Reporting & Data System (Lung-RADS) version 2022, and Ovarian-Adnexal Reporting & Data System (O-RADS) magnetic resonance imaging, meticulously prepared by board-certified radiologists. Each report underwent 6 assessments. Two blinded reviewers assessed the chatbots’ response at patient-level RADS categorization and overall ratings. The agreement across repetitions was assessed using Fleiss κ. Results: Claude-2 achieved the highest accuracy in overall ratings with few-shot prompts and guideline PDFs (prompt-2), attaining 57% (17/30) average accuracy over 6 runs and 50% (15/30) accuracy with k-pass voting. Without prompt engineering, all chatbots performed poorly. The introduction of a structured exemplar prompt (prompt-1) increased the accuracy of overall ratings for all chatbots. Providing prompt-2 further improved Claude-2’s performance, an enhancement not replicated by GPT-4. The interrun agreement was substantial for Claude-2 (k=0.66 for overall rating and k=0.69 for RADS categorization), fair for GPT-4 (k=0.39 for both), and fair for GPT-3.5 (k=0.21 for overall rating and k=0.39 for RADS categorization). All chatbots showed significantly higher accuracy with LI-RADS version 2018 than with Lung-RADS version 2022 and O-RADS (P<.05); with prompt-2, Claude-2 achieved the highest overall rating accuracy of 75% (45/60) in LI-RADS version 2018. Conclusions: When equipped with structured prompts and guideline PDFs, Claude-2 demonstrated potential in assigning RADS categories to radiology cases according to established criteria such as LI-RADS version 2018. However, the current generation of chatbots lags in accurately categorizing cases based on more recent RADS criteria.


  • Open Access

    Evaluating Large Language Models for Automated Reporting and Data Systems Categorization: Cross-Sectional Study (Preprint)

    by Qingxia Wu, Huali Li, Yan Wang, Yan Bai, Yaping Wu, Xuan Yu, Xiaodong Li, Pei Dong, Jon Xue, Dinggang Shen, et al.

    Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Large language models show promise for improving radiology workflows, but their performance on structured radiological tasks such as Reporting and Data Systems (RADS) categorization remains unexplored. Objective: This study aims to evaluate 3 large language model chatbots—Claude-2, GPT-3.5, and GPT-4—on assigning RADS categories to radiology reports and assess the impact of different prompting strategies. Methods: This cross-sectional study compared 3 chatbots using 30 radiology reports (10 per RADS criteria), using a 3-level prompting strategy: zero-shot, few-shot, and guideline PDF-informed prompts. The cases were grounded in Liver Imaging Reporting & Data System (LI-RADS) version 2018, Lung CT (computed tomography) Screening Reporting & Data System (Lung-RADS) version 2022, and Ovarian-Adnexal Reporting & Data System (O-RADS) magnetic resonance imaging, meticulously prepared by board-certified radiologists. Each report underwent 6 assessments. Two blinded reviewers assessed the chatbots’ response at patient-level RADS categorization and overall ratings. The agreement across repetitions was assessed using Fleiss κ. Results: Claude-2 achieved the highest accuracy in overall ratings with few-shot prompts and guideline PDFs (prompt-2), attaining 57% (17/30) average accuracy over 6 runs and 50% (15/30) accuracy with k-pass voting. Without prompt engineering, all chatbots performed poorly. The introduction of a structured exemplar prompt (prompt-1) increased the accuracy of overall ratings for all chatbots. Providing prompt-2 further improved Claude-2’s performance, an enhancement not replicated by GPT-4. The interrun agreement was substantial for Claude-2 (k=0.66 for overall rating and k=0.69 for RADS categorization), fair for GPT-4 (k=0.39 for both), and fair for GPT-3.5 (k=0.21 for overall rating and k=0.39 for RADS categorization). All chatbots showed significantly higher accuracy with LI-RADS version 2018 than with Lung-RADS version 2022 and O-RADS (P<.05); with prompt-2, Claude-2 achieved the highest overall rating accuracy of 75% (45/60) in LI-RADS version 2018. Conclusions: When equipped with structured prompts and guideline PDFs, Claude-2 demonstrated potential in assigning RADS categories to radiology cases according to established criteria such as LI-RADS version 2018. However, the current generation of chatbots lags in accurately categorizing cases based on more recent RADS criteria.

  • Multi-Atlas Segmentation of Anatomical Brain Structures Using Hierarchical Hypergraph Learning

    by Pei Dong, Yanrong Guo, Yue Gao, Peipeng Liang, Yonghong Shi, Guorong Wu

    Accurate segmentation of anatomical brain structures is crucial for many neuroimaging applications, e.g., early brain development studies and the study of imaging biomarkers of neurodegenerative diseases. Although multi-atlas segmentation (MAS) has achieved many successes in the medical imaging area, this approach encounters limitations in segmenting anatomical structures associated with poor image contrast. To address this issue, we propose a new MAS method that uses a hypergraph learning framework to model the complex subject-within and subject-to-atlas image voxel relationships and propagate the label on the atlas image to the target subject image. To alleviate the low-image contrast issue, we propose two strategies equipped with our hypergraph learning framework. First, we use a hierarchical strategy that exploits high-level context features for hypergraph construction. Because the context features are computed on the tentatively estimated probability maps, we can ultimately turn the hypergraph learning into a hierarchical model. Second, instead of only propagating the labels from the atlas images to the target subject image, we use a dynamic label propagation strategy that can gradually use increasing reliably identified labels from the subject image to aid in predicting the labels on the difficult-to-label subject image voxels. Compared with the state-of-the-art label fusion methods, our results show that the hierarchical hypergraph learning framework can substantially improve the robustness and accuracy in the segmentation of anatomical brain structures with low image contrast from magnetic resonance (MR) images.


  • Open Access

    Fast Groupwise Registration Using Multi-Level and Multi-Resolution Graph Shrinkage

    by Pei Dong, Xiaohuan Cao, Pew-Thian Yap, Dinggang Shen

    Groupwise registration aligns a set of images to a common space. It can however be inefficient and ineffective when dealing with datasets with significant anatomical variations. To mitigate these problems, we propose a groupwise registration framework based on hierarchical multi-level and multi-resolution shrinkage of a graph set. First, to deal with datasets with complex inhomogeneous image distributions, we divide the images hierarchically into multiple clusters. Since the images in each cluster have similar appearances, they can be registered effectively. Second, we employ a multi-resolution strategy to reduce computational cost. Experimental results on two public datasets show that our proposed method yields state-of-the-art registration accuracy with significantly reduced computational time.


  • Open Access

    Efficient Groupwise Registration of MR Brain Images via Hierarchical Graph Set Shrinkage

    by Pei Dong, Xiaohuan Cao, Pew-Thian Yap, Dinggang Shen

    Accurate and efficient groupwise registration is important for population analysis. Current groupwise registration methods suffer from high computational cost, which hinders their application to large image datasets. To alleviate the computational burden while delivering accurate groupwise registration result, we propose to use a hierarchical graph set to model the complex image distribution with possibly large anatomical variations, and then turn the groupwise registration problem as a series of simple-to-solve graph shrinkage problems. Specifically, first, we divide the input images into a set of image clusters hierarchically, where images within each image cluster have similar anatomical appearances whereas images falling into different image clusters have varying anatomical appearances. After clustering, two types of graphs, i.e., intra-graph and inter-graph, are employed to hierarchically model the image distribution both within and across the image clusters. The constructed hierarchical graph set divides the registration problem of the whole image set into a series of simple-to-solve registration problems, where the entire registration process can be solved accurately and efficiently. The final deformation pathway of each image to the estimated population center can be obtained by composing each part of the deformation pathway along the hierarchical graph set. To evaluate our proposed method, we performed registration of a hundred of brain images with large anatomical variations. The results indicate that our method yields significant improvement in registration performance over state-of-the-art groupwise registration methods.


  • Scalable joint segmentation and registration framework for infant brain images

    by Pei Dong, Li Wang, Weili Lin, Dinggang Shen, Guorong Wu


  • Open Access

    Multi-Atlas Based Segmentation of Brainstem Nuclei from MR Images by Deep Hyper-Graph Learning

    by Pei Dong, Yangrong Guo, Yue Gao, Peipeng Liang, Yonghong Shi, Qian Wang, Dinggang Shen, Guorong Wu

    Accurate segmentation of brainstem nuclei (red nucleus and substantia nigra) is very important in various neuroimaging applications such as deep brain stimulation and the investigation of imaging biomarkers for Parkinson’s disease (PD). Due to iron deposition during aging, image contrast in the brainstem is very low in Magnetic Resonance (MR) images. Hence, the ambiguity of patch-wise similarity makes the recently successful multi-atlas patch-based label fusion methods have difficulty to perform as competitive as segmenting cortical and sub-cortical regions from MR images. To address this challenge, we propose a novel multi-atlas brainstem nuclei segmentation method using deep hyper-graph learning. Specifically, we achieve this goal in three-fold. First, we employ hyper-graph to combine the advantage of maintaining spatial coherence from graph-based segmentation approaches and the benefit of harnessing population priors from multi-atlas based framework. Second, besides using low-level image appearance, we also extract high-level context features to measure the complex patch-wise relationship. Since the context features are calculated on a tentatively estimated label probability map, we eventually turn our hyper-graph learning based label propagation into a deep and self-refining model. Third, since anatomical labels on some voxels (usually located in uniform regions) can be identified much more reliably than other voxels (usually located at the boundary between two regions), we allow these reliable voxels to propagate their labels to the nearby difficult-to-label voxels. Such hierarchical strategy makes our proposed label fusion method deep and dynamic. We evaluate our proposed label fusion method in segmenting substantia nigra (SN) and red nucleus (RN) from 3.0 T MR images, where our proposed method achieves significant improvement over the state-of-the-art label fusion methods.


  • Open Access

    QUANTIFICATION OF THE 3D MORPHOLOGY OF THE BONE CELL NETWORK FROM SYNCHROTRON MICRO-CT IMAGES

    by Pei Dong, Alexandra Pacureanu, Maria Alejandra Zuluaga, Cécile Olivier, Quentin Grimal, Françoise Peyrin

    In the context of bone diseases research, recent works have highlighted the crucial role of the osteocyte system. This system, hosted in the lacuno-canalicular network (LCN), plays a key role in the bone remodeling process. However, few data are available on the LCN due to the limitations of current microscopy techniques, and have mainly only been obtained from 2D histology sections. Here we present, for the first time, an automatic method to quantify the LCN in 3D from synchrotron radiation micro-tomography images. After segmentation of the LCN, two binary images are generated, one of lacunae (hosting the cell body) and one of canaliculi (small channels linking the lacunae). The binary image of lacunae is labeled, and for each object, lacunar descriptors are extracted after calculating the second order moments and the intrinsic volumes. Furthermore, we propose a specific method to quantify the ramification of canaliculi around each lacuna. To this aim, a signature of the numbers of canaliculi at different distances from the lacunar surface is estimated through the calculation of topological parameters. The proposed method was applied to the 3D SR micro-CT image of a human femoral mid-diaphysis bone sample. Statistical results are reported on 399 lacunae and their surrounding canaliculi.